I think we are talking about different things.
whant I want to do is predict the dependent values just using part of
the effects from
left hand side of the equation while not changing the values of the
effects estimated
from the "whole" model.
It sounds like the same thing to me. You have estimated a model that
includes gender. You now want to pretend that the effects of gender were 0
while all the other effects were the same. I'm not sure why you would want
to do that. As you've already noted, the other coefficients change when
gender is dropped from the model. Why not just use the estimates from a
model where gender is excluded?
